Feather the Throttle: Revisiting Visual Token Pruning for Vision-Language Model Acceleration

Recent works on accelerating Vision-Language Models show that strong performance can be maintained across a variety of vision-language tasks despite highly compressing visual information. In this work, we examine the popular acceleration approach of early pruning of visual tokens inside the language model and find that its strong performance across many tasks is not due to an exceptional ability to compress visual information, but rather the benchmarks' limited ability to assess fine-grained visual capabilities. Namely, we demonstrate a core issue with the acceleration approach where most tokens towards the top of the image are pruned away. Yet, this issue is only reflected in performance for a small subset of tasks such as localization. For the other evaluated tasks, strong performance is maintained with the flawed pruning strategy. Noting the limited visual capabilities of the studied acceleration technique, we propose FEATHER (Fast and Effective Acceleration wiTH Ensemble cRiteria), a straightforward approach that (1) resolves the identified issue with early-layer pruning, (2) incorporates uniform sampling to ensure coverage across all image regions, and (3) applies pruning in two stages to allow the criteria to become more effective at a later layer while still achieving significant speedup through early-layer pruning. With comparable computational savings, we find that FEATHER has more than $5\times$ performance improvement on the vision-centric localization benchmarks compared to the original acceleration approach.

Apollo: An Exploration of Video Understanding in Large Multimodal Models

Despite the rapid integration of video perception capabilities into Large Multimodal Models (LMMs), the underlying mechanisms driving their video understanding remain poorly understood. Consequently, many design decisions in this domain are made without proper justification or analysis. The high computational cost of training and evaluating such models, coupled with limited open research, hinders the development of video-LMMs. To address this, we present a comprehensive study that helps uncover what effectively drives video understanding in LMMs. We begin by critically examining the primary contributors to the high computational requirements associated with video-LMM research and discover Scaling Consistency, wherein design and training decisions made on smaller models and datasets (up to a critical size) effectively transfer to larger models. Leveraging these insights, we explored many video-specific aspects of video-LMMs, including video sampling, architectures, data composition, training schedules, and more. For example, we demonstrated that fps sampling during training is vastly preferable to uniform frame sampling and which vision encoders are the best for video representation. Guided by these findings, we introduce Apollo, a state-of-the-art family of LMMs that achieve superior performance across different model sizes. Our models can perceive hour-long videos efficiently, with Apollo-3B outperforming most existing $7$B models with an impressive 55.1 on LongVideoBench. Apollo-7B is state-of-the-art compared to 7B LMMs with a 70.9 on MLVU, and 63.3 on Video-MME.

DeforHMR: Vision Transformer with Deformable Cross-Attention for 3D Human Mesh Recovery

Human Mesh Recovery (HMR) is an important yet challenging problem with applications across various domains including motion capture, augmented reality, and biomechanics. Accurately predicting human pose parameters from a single image remains a challenging 3D computer vision task. In this work, we introduce DeforHMR, a novel regression-based monocular HMR framework designed to enhance the prediction of human pose parameters using deformable attention transformers. DeforHMR leverages a novel query-agnostic deformable cross-attention mechanism within the transformer decoder to effectively regress the visual features extracted from a frozen pretrained vision transformer (ViT) encoder. The proposed deformable cross-attention mechanism allows the model to attend to relevant spatial features more flexibly and in a data-dependent manner. Equipped with a transformer decoder capable of spatially-nuanced attention, DeforHMR achieves state-of-the-art performance for single-frame regression-based methods on the widely used 3D HMR benchmarks 3DPW and RICH. By pushing the boundary on the field of 3D human mesh recovery through deformable attention, we introduce an new, effective paradigm for decoding local spatial information from large pretrained vision encoders in computer vision.

Motion Diffusion-Guided 3D Global HMR from a Dynamic Camera

Motion capture technologies have transformed numerous fields, from the film and gaming industries to sports science and healthcare, by providing a tool to capture and analyze human movement in great detail. The holy grail in the topic of monocular global human mesh and motion reconstruction (GHMR) is to achieve accuracy on par with traditional multi-view capture on any monocular videos captured with a dynamic camera, in-the-wild. This is a challenging task as the monocular input has inherent depth ambiguity, and the moving camera adds additional complexity as the rendered human motion is now a product of both human and camera movement. Not accounting for this confusion, existing GHMR methods often output motions that are unrealistic, e.g. unaccounted root translation of the human causes foot sliding. We present DiffOpt, a novel 3D global HMR method using Diffusion Optimization. Our key insight is that recent advances in human motion generation, such as the motion diffusion model (MDM), contain a strong prior of coherent human motion. The core of our method is to optimize the initial motion reconstruction using the MDM prior. This step can lead to more globally coherent human motion. Our optimization jointly optimizes the motion prior loss and reprojection loss to correctly disentangle the human and camera motions. We validate DiffOpt with video sequences from the Electromagnetic Database of Global 3D Human Pose and Shape in the Wild (EMDB) and Egobody, and demonstrate superior global human motion recovery capability over other state-of-the-art global HMR methods most prominently in long video settings.

Zero-shot Action Localization via the Confidence of Large Vision-Language Models

Precise action localization in untrimmed video is vital for fields such as professional sports and minimally invasive surgery, where the delineation of particular motions in recordings can dramatically enhance analysis. But in many cases, large scale datasets with video-label pairs for localization are unavailable, limiting the opportunity to fine-tune video-understanding models. Recent developments in large vision-language models (LVLM) address this need with impressive zero-shot capabilities in a variety of video understanding tasks. However, the adaptation of image-based LVLMs, with their powerful visual question answering capabilities, to action localization in long-form video is still relatively unexplored. To this end, we introduce a true ZEro-shot Action Localization method (ZEAL). Specifically, we leverage the built-in action knowledge of a large language model (LLM) to inflate actions into highly-detailed descriptions of the archetypal start and end of the action. These descriptions serve as queries to LVLM for generating frame-level confidence scores which can be aggregated to produce localization outputs. The simplicity and flexibility of our method lends it amenable to more capable LVLMs as they are developed, and we demonstrate remarkable results in zero-shot action localization on a challenging benchmark, without any training.

How to Build the Virtual Cell with Artificial Intelligence: Priorities and Opportunities

The cell is arguably the smallest unit of life and is central to understanding biology. Accurate modeling of cells is important for this understanding as well as for determining the root causes of disease. Recent advances in artificial intelligence (AI), combined with the ability to generate large-scale experimental data, present novel opportunities to model cells. Here we propose a vision of AI-powered Virtual Cells, where robust representations of cells and cellular systems under different conditions are directly learned from growing biological data across measurements and scales. We discuss desired capabilities of AI Virtual Cells, including generating universal representations of biological entities across scales, and facilitating interpretable in silico experiments to predict and understand their behavior using Virtual Instruments. We further address the challenges, opportunities and requirements to realize this vision including data needs, evaluation strategies, and community standards and engagement to ensure biological accuracy and broad utility. We envision a future where AI Virtual Cells help identify new drug targets, predict cellular responses to perturbations, as well as scale hypothesis exploration. With open science collaborations across the biomedical ecosystem that includes academia, philanthropy, and the biopharma and AI industries, a comprehensive predictive understanding of cell mechanisms and interactions is within reach.

Continuous Perception Benchmark

Humans continuously perceive and process visual signals. However, current video models typically either sample key frames sparsely or divide videos into chunks and densely sample within each chunk. This approach stems from the fact that most existing video benchmarks can be addressed by analyzing key frames or aggregating information from separate chunks. We anticipate that the next generation of vision models will emulate human perception by processing visual input continuously and holistically. To facilitate the development of such models, we propose the Continuous Perception Benchmark, a video question answering task that cannot be solved by focusing solely on a few frames or by captioning small chunks and then summarizing using language models. Extensive experiments demonstrate that existing models, whether commercial or open-source, struggle with these tasks, indicating the need for new technical advancements in this direction.

Video-STaR: Self-Training Enables Video Instruction Tuning with Any Supervision

The performance of Large Vision Language Models (LVLMs) is dependent on the size and quality of their training datasets. Existing video instruction tuning datasets lack diversity as they are derived by prompting large language models with video captions to generate question-answer pairs, and are therefore mostly descriptive. Meanwhile, many labeled video datasets with diverse labels and supervision exist - however, we find that their integration into LVLMs is non-trivial. Herein, we present Video Self-Training with augmented Reasoning (Video-STaR), the first video self-training approach. Video-STaR allows the utilization of any labeled video dataset for video instruction tuning. In Video-STaR, an LVLM cycles between instruction generation and finetuning, which we show (I) improves general video understanding and (II) adapts LVLMs to novel downstream tasks with existing supervision. During generation, an LVLM is prompted to propose an answer. The answers are then filtered only to those that contain the original video labels, and the LVLM is then re-trained on the generated dataset. By only training on generated answers that contain the correct video labels, Video-STaR utilizes these existing video labels as weak supervision for video instruction tuning. Our results demonstrate that Video-STaR-enhanced LVLMs exhibit improved performance in (I) general video QA, where TempCompass performance improved by 10%, and (II) on downstream tasks, where Video-STaR improved Kinetics700-QA accuracy by 20% and action quality assessment on FineDiving by 15%.

μ-Bench: A Vision-Language Benchmark for Microscopy Understanding

Recent advances in microscopy have enabled the rapid generation of terabytes of image data in cell biology and biomedical research. Vision-language models (VLMs) offer a promising solution for large-scale biological image analysis, enhancing researchers' efficiency, identifying new image biomarkers, and accelerating hypothesis generation and scientific discovery. However, there is a lack of standardized, diverse, and large-scale vision-language benchmarks to evaluate VLMs' perception and cognition capabilities in biological image understanding. To address this gap, we introduce {\mu}-Bench, an expert-curated benchmark encompassing 22 biomedical tasks across various scientific disciplines (biology, pathology), microscopy modalities (electron, fluorescence, light), scales (subcellular, cellular, tissue), and organisms in both normal and abnormal states. We evaluate state-of-the-art biomedical, pathology, and general VLMs on {\mu}-Bench and find that: i) current models struggle on all categories, even for basic tasks such as distinguishing microscopy modalities; ii) current specialist models fine-tuned on biomedical data often perform worse than generalist models; iii) fine-tuning in specific microscopy domains can cause catastrophic forgetting, eroding prior biomedical knowledge encoded in their base model. iv) weight interpolation between fine-tuned and pre-trained models offers one solution to forgetting and improves general performance across biomedical tasks. We release {\mu}-Bench under a permissive license to accelerate the research and development of microscopy foundation models.

Why are Visually-Grounded Language Models Bad at Image Classification?

Image classification is one of the most fundamental capabilities of machine vision intelligence. In this work, we revisit the image classification task using visually-grounded language models (VLMs) such as GPT-4V and LLaVA. We find that existing proprietary and public VLMs, despite often using CLIP as a vision encoder and having many more parameters, significantly underperform CLIP on standard image classification benchmarks like ImageNet. To understand the reason, we explore several hypotheses concerning the inference algorithms, training objectives, and data processing in VLMs. Our analysis reveals that the primary cause is data-related: critical information for image classification is encoded in the VLM's latent space but can only be effectively decoded with enough training data. Specifically, there is a strong correlation between the frequency of class exposure during VLM training and instruction-tuning and the VLM's performance in those classes; when trained with sufficient data, VLMs can match the accuracy of state-of-the-art classification models. Based on these findings, we enhance a VLM by integrating classification-focused datasets into its training, and demonstrate that the enhanced classification performance of the VLM transfers to its general capabilities, resulting in an improvement of 11.8% on the newly collected ImageWikiQA dataset.